ncnn

1
// yala is pleased to support the open source community by making ncnn available.
2
//
3
//
4
// Copyright (C) 2022 yala <zhaojunchao@loongson.cn>;<junchao82@qq.com>. All rights reserved.
5
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
6
// in compliance with the License. You may obtain a copy of the License at
7
//
8
// https://opensource.org/licenses/BSD-3-Clause
9
//
10
// Unless required by applicable law or agreed to in writing, software distributed
11
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
12
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
13
// specific language governing permissions and limitations under the License.
14

15
#include "sigmoid_loongarch.h"
16

17
#if __loongarch_sx
18
#include <lsxintrin.h>
19
#include "lsx_mathfun.h"
20
#endif // __loongarch_sx
21

22
#include "loongarch_usability.h"
23

24
namespace ncnn {
25

26
Sigmoid_loongarch::Sigmoid_loongarch()
27
{
28
#if __loongarch_sx
29
    support_packing = true;
30
#endif
31
}
32

33
int Sigmoid_loongarch::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
34
{
35
    int w = bottom_top_blob.w;
36
    int h = bottom_top_blob.h;
37
    int d = bottom_top_blob.d;
38
    int channels = bottom_top_blob.c;
39
    int elempack = bottom_top_blob.elempack;
40
    int size = w * h * d * elempack;
41

42
    #pragma omp parallel for num_threads(opt.num_threads)
43
    for (int q = 0; q < channels; q++)
44
    {
45
        float* ptr = bottom_top_blob.channel(q);
46

47
        int i = 0;
48
#if __loongarch_sx
49
        __m128 _one = (__m128)__lsx_vreplfr2vr_s(1.f);
50
        for (; i + 3 < size; i += 4)
51
        {
52
            __builtin_prefetch(ptr + 16);
53
            __m128 _p = (__m128)__lsx_vld(ptr, 0);
54
            _p = (__m128)__lsx_vbitrevi_w((__m128i)_p, 31);
55
            _p = exp_ps(_p);
56
            _p = __lsx_vfadd_s(_p, _one);
57
            __m128 _outp = __lsx_vfdiv_s(_one, _p);
58
            __lsx_vst(_outp, ptr, 0);
59

60
            ptr += 4;
61
        }
62
#endif // __loongarch_sx
63
        for (; i < size; i++)
64
        {
65
            *ptr = 1.f / (1.f + exp(-*ptr));
66

67
            ptr++;
68
        }
69
    }
70

71
    return 0;
72
}
73

74
} // namespace ncnn
75